The mold book writes itself while the mold is built

The problem

Every mold is a unique story, and its evidence is scattered.

Automotive documentation requirements are not negotiable, but the evidence is generated scattered and gets ordered by hand and out of time. In a project shop the problem is worse: there is no procedure repeated a hundred times, there are a hundred unique stories. Every mold had its own engineering changes, its own tryout loops and its own repairs. The typical result is an enormous effort concentrated exactly when the team should be pushing the next molds out — and a gap the auditor eventually finds.

• The day a customer asks for the full history of a mold delivered two years ago, someone spends days rebuilding it across the system, a network folder, photos on phones and three people's memory.
• Automotive documentation requirements are not negotiable, but the evidence is generated scattered and gets ordered by hand and out of time.
• In a project shop it is worse: there is no procedure repeated a hundred times, there are a hundred unique stories, each with its own engineering changes, tryout loops and repairs.
• The typical result is an enormous effort concentrated exactly when the team should be pushing the next molds out — and a gap the auditor eventually finds.

How it fits the IRIS system

Agents on top of central memory — nothing new has to be captured.

Agents on top of central memory. Nothing new has to be captured: it is all in the system already from the previous cases.

It is all in the system already from the previous cases: job travelers, panel data, alerts and engineering changes, bench decisions, validations, dimensional reports, heats and certificates, machine hours, tryout defects and blocked closings. The Agents order that evidence by the requirements of each applicable standard, publish a browsable dossier, and the quality manager reviews and signs. Gaps are flagged during the build, not on audit day, while the detail is still on the bench and the missing measurement can still be taken.

• The Agents assemble per mold the job travelers, the machine panel data, the alerts and engineering changes, the bench decisions, the validations, the dimensional reports, the heats and certificates, the machine hours, the tryout defects and the blocked closings.
• They order that evidence by the requirements of each applicable automotive standard.
• They publish a browsable dossier; the quality manager reviews and signs.
• The auditor or the customer logs in and downloads the package for the mold or period requested.

And the most valuable part: gaps are flagged by the system during the build, while they can still be closed.

See the full IRIS architecture →

Before and after

Today's mold history versus the living mold book

Aspect	Today	With iLEAN Agents
History of a delivered mold	Days of document archaeology	The dossier already exists
Where the evidence is	System, folder, phones and memory	One package per mold
Ordering by standard	By hand, before the audit	By the Agents, as it happens
A documentation gap	Found by the auditor	Flagged by the system in time
Heat, revision and tryout history	Three different searches	One lookup by mold
The customer's access	An email with attachments	A scoped download of the mold requested

from days of document archaeology per audit or complaint → the dossier already exists. From gaps found by the auditor → gaps flagged by the system in time.

Impact estimate

Impact estimate — to be validated with your numbers.

The block below is an estimate to be validated against your plant's actual data. We put it forward so the committee has an order of magnitude; we refine it during the assessment.

• Estimated payback 5-11 months from quality and engineering time recovered.
• Reduced nonconformity risk, which in automotive is paid in competitive position at the next award.
• The effort no longer concentrates in the weeks when the team should be pushing the next molds out.
• And a complaint on a delivered mold is answered from the dossier, not from three people's memory.

estimated payback 5-11 months from quality and engineering time recovered, plus reduced nonconformity risk — which in automotive is paid in competitive position at the next award. *Estimate to validate*.

And the fair question from the production manager

"Isn't this just a document manager with a nicer name?" — the difference is direction. A document manager stores what somebody decides to upload, after the fact, and inherits the same gaps. Here the evidence enters at the moment it happens, signed at the bench, and what the Agents do is reorganize it by whichever standard asks. Each piece of evidence was read as an anchored task, where the best models drop below 1.5% error [1], and validated by a person before it became part of the mold's history.

^[1] OpenAI paper "Why Language Models Hallucinate", 2025 — on the reliability of AI in anchored tasks.